[Interest] Some questions on QSettings, qmlRegisterType() and setContextProperty

Alan Alpert 416365416c at gmail.com
Sat Aug 10 00:04:19 CEST 2013


On Fri, Aug 9, 2013 at 8:12 AM, Jayesh Bhoot <mister.bhoot at gmail.com> wrote:
> I will try explaining my confusion through the application I am currently
> developing.
>
> My application (based on Qt5.1 + Qt Quick Controls) interacts with
> Facebook API to manage a Facebook Page. I am trying to keep the QML code
> (for UI) as separate as possible from the C++ core.
>
> Now, an OAuth2 implementation is required to be able to interact with
> Facebook API. For that, I have a C++ OAuth2 class, the constructor of
> which has the following signature:
>
> OAuth2::OAuth2(QString appId, QString redirectUrl, QStringList
> permissions);
>
> Now, as the OAuth process requires a browser, I have also implemented an
> OAuthBrowser.qml, which uses OAuth2 to complete an authorization.
>
> I have the following options to expose OAuth2 class to OAuth2Browser:
>
> 1. Instantiate OAuth2 and use setContextProperty() to expose the instance
> to OAuth2Browser. However, this means my C++ code has to deal with the UI
> code. The more baffling issue is that OAuth2Browser is a secondary window.
> When a user clicks on a "Authorize" window on the MainWindow, then an
> AppController C++ object (connected to MainWindow) will launch the
> OAuth2Browser window. Thus, the instantiation code of OAuth2Browser would
> go deep down inside a AppController method. It would have been good if
> only main.cpp had to deal with the window creation.
>
> 2. Use qmlRegisterType(). In this case, I can't pass parameters to the
> constructor. So, I will have to implement an init() method that would
> initialize an OAuth2 object. Then, I would call this init() method in
> OAuth2Browser's Component.onCompleted() method.
> However, in this approach, I will have to expose QSettings to the UI code
> - QML window, so that the required parameters to init() method can be
> retrieved. I have huge skepticism on whether directly exposing application
> settings to QML UI is a good idea.
>
> 3. Implicitly use QSettings within the OAuth2 constructor. This way, I
> won't have to pass any parameters, and I would be able to use
> qmlRegisterType(). However, this means I am doing some magic stuff "behind
> the curtains". Instead of explicitly passing QSettings instance, I am
> using it wherever the hell I want to, thus hiding the initialization
> detail from public API.
>
> An alternative based on the 3rd option was advised on IRC - use an
> initFromSettings() type of method to initialize an instance if no
> parameter is passed to the constructor. That way, the initialization is
> not hidden, and initFromSettings() can confidently use QSettings within
> itself. Now, I can happily use qmlRegisterType() to instantiate OAuth2 in
> QML.
>
> So, what is the better approach?

2., but calling init in C++ componentComplete() (derive from
QQmlParserStatus as well) automatically, taking the input from
properties set in QML.

> Also,
> 1. Is exposing QSettings directly to QML UI a good idea?

Directly? Not a good idea, but possible. The new Qt.labs.settings API
aims to address this deficiency:
https://codereview.qt-project.org/#change,59149

> 2. I personally prefer qmlRegisterType() to setContextProperty() - that
> way, the lifetime of a registered class's instance is maintained solely by
> QML. However, the former is less likely to be used due to the lack of
> support of parameterized constructors, unless some form of init() is used
> explicitly for initialization. Is that a good design?

qmlRegisterType is better than setContextProperty, because that way
the flow is controlled from QML (and a lack of constructor arguments
is correct for declarative design, you can do additional property
based initialization in componentComplete). Good QML design has C++
expose functionality to QML but lets QML control the entire UI flow
(including window creation/handling).

>
> I apologise in advance for an excruciatingly long post. But I thought it
> best to ask here.

Sorry for the excruciatingly short reply ;)

--
Alan Alpert



More information about the Interest mailing list